The α-β algorithm
| ALPHA-BETA-SEARCH adds alpha
(max) and beta (min) values to the MINIMAX algorithm. In a node at a MAX level (an odd level), we
know that its parent, a MIN node, will never choose this MAX node if there is
another previously evaluated node with a known lower utility value. This is represented by beta (the value of
the lowest-value choice along the path to that MIN node). In this case the search can be pruned and
we can save execution time by not evaluating the remaining leaves of the
tree. |
|
| In a node at the MIN level
(an even level), we have a similar situation that uses alpha to decide
whether to prune the node or not.
Figure 6.5d (pg. 168) shows where a pruning action in the MIN-VALUE
function kicks in. In Figure 6.5,
pruning in MAX-VALUE doesn’t occur.
The pruning is brought about by the return statement in the
SUCCESSORS function in the MIN- and MAX-VALUE procedures. |
|